Medical diagnostic aid and method

1. A method for assisting medical personnel in performing a diagnosis, the method comprising an initialisation step and a diagnostic step,

wherein the initialisation step comprises:

receiving a plurality of input medical texts tagged with potential medical codes;

curating and validating the plurality of input medical texts to output a subset of medical texts that are validated and tagged with medical codes; and

using the subset of tagged and validated medical texts to model the associations between the medical texts and the medical codes, and generating a medical text learning model based on the associations, wherein the associations between the medical texts and the medical codes for the tagged and validated medical texts are modelled using a Labelled Latent Dirichlet Allocation topic model, and wherein the medical text learning model uses a convolutional neural network, and

wherein the diagnostic step comprises:

inputting a specimen text relating to a patient into the medical text learning model;

processing the specimen text using the medical text learning model;

identifying suggested medical codes based on the specimen text; and

outputting diagnoses, symptoms and treatments linked to the suggested medical codes for assisting medical personnel in providing a diagnosis for the patient.

2. The method of claim 1 , wherein the curation and validation of the plurality of input medical texts comprises selecting the subset of medical texts based on at least one of:

the number of potential medical codes associated with text;

the language text is written in; and

the length of text.

3. The method of claim 1 , wherein the curation and validation involves input from medical personnel.

4. The method of claim 3 , wherein the curation and validation comprises at least an initial stage and a final stage, wherein medical personnel involved in the initial stage are less experienced than medical personnel involved in the final stage, and wherein the medical personnel involved in the final stage select medical personnel for involvement in the initial stage.

5. The method of claim 1 , wherein the plurality of input medical texts are natural language texts.

6. The method of claim 5 , wherein the natural language texts are notes relating to a patient consultation.

7. The method of claim 1 , further comprising generating the plurality of input texts tagged with potential medical codes by:

receiving an unstructured input;

splitting the unstructured input into a plurality of logical components, and detecting medical terms in the plurality of logical components;

receiving a medical classification hierarchy of medical standard codes in the form of a knowledge graph, and semantically annotating the knowledge graph with synonyms of medical terms used in the medical standard codes;

analysing the medical terms detected in the plurality of logical components by a text analyser and parser, generating a list of potential matching medical standard codes for each of the medical terms, comparing the lists of potential matching medical standard codes, and outputting top matching medical standard codes based on the comparison; and

inputting the unstructured input tagged with the top matching medical standard codes as one of the plurality of input medical texts.

8. The method of claim 7 , wherein the step of inputting of the specimen text comprises accessing a medical record linked to a patient that is the subject of the specimen text, and the outputting diagnoses, symptoms and treatments linked to the suggested medical codes for assisting medical personnel in providing a diagnosis for the patient further comprises outputting the medical record.

9. The method of claim 8 , further comprising updating the medical record with the output diagnoses, symptoms and treatments.

10. A medical diagnostic aid for assisting medical personnel in performing a diagnosis, the diagnostic aid comprising:

a first receiver configured to receive a plurality of input medical texts tagged with potential medical codes;

a data curator configured to curate and validate the plurality of input medical texts, and to output a subset of medical texts that are validated and tagged with medical codes;

a classification model engine configured to use the subset of tagged and validated medical texts to model the associations between the medical texts and the medical codes, and to generate a medical text learning model based on the associations, wherein the associations between the medical texts and the medical codes for the tagged and validated medical texts are modelled using a Labelled Latent Dirichlet Allocation topic model, and wherein the medical text learning model uses a convolutional neural network;

an inputter configured to input a specimen text relating to a patient into the medical text learning model;

a specimen text processor configured to process the specimen text using the medical text learning model, and identify suggested medical codes based on the specimen text; and

an outputter configured to output diagnoses, symptoms and treatments linked to the suggested medical codes for assisting medical personnel in providing a diagnosis for the patient.

11. The method of claim 7 , further comprising generating a single list of top matching medical standard code from the plurality of logical components.

12. The medical diagnostic aid of claim 10 , further comprising:

a second receiver configured to receive an unstructured input;

an analyser and parser configured to split the unstructured input into a plurality of logical components, and to detect medical terms in the plurality of logical components;

a mapping engine configured to receive a medical classification hierarchy of medical standard codes in the form of a knowledge graph, and semantically annotate the knowledge graph with synonyms of medical terms used in the medical standard codes;

an automatic coding solver configured to analyse the medical terms detected in the plurality of logical components by the analyser and parser, to generate a list of potential matching medical standard codes for each of the medical terms, to compare the lists of potential matching medical standard codes, and to output top matching medical standard codes based on the comparison; and

an enrichment engine configured to enrich the unstructured input with the top matching medical standard codes, and to output the unstructured input tagged with the top matching medical standard codes as one of the plurality of input medical texts received by the first receiver.

13. The medical diagnostic aid of claim 12 , wherein the automatic coding solver is further configured to generate a single list of top matching medical standard code from the plurality of logical components.

14. A computer readable medium comprising code which, when executed by a computer, causes the computer to execute the method of claim 1 .

15. A diagnostic device, comprising:

initialization circuitry configured to

receive a plurality of input medical texts tagged with potential medical codes;

curate and validate the plurality of input medical texts to output a subset of medical texts that are validated and tagged with medical codes;

use the subset of tagged and validated medical texts to model the associations between the medical texts and the medical codes, and generate a medical text learning model based on the associations, wherein the associations between the medical texts and the medical codes for the tagged and validated medical texts are modelled using a Labelled Latent Dirichlet Allocation topic model, and wherein the medical text learning model uses a convolutional neural network; and

diagnostic circuitry configured to

receive a specimen text relating to a patient into the medical text learning model;

process the specimen text using the medical text learning model;

identify suggested medical codes based on the specimen text; and

output diagnoses, symptoms and treatments linked to the suggested medical codes for assisting medical personnel in providing a diagnosis for the patient.

16. The diagnostic device of claim 15 , further comprising:

tagging circuitry configured to

split an unstructured input into a plurality of logical components, and to detect medical terms in the plurality of logical components;

receive a medical classification hierarchy of medical standard codes in the form of a knowledge graph, and semantically annotate the knowledge graph with synonyms of medical terms used in the medical standard codes;

analyse the medical terms detected in the plurality of logical components by the analyser and parser, to generate a list of potential matching medical standard codes for each of the medical terms, to compare the lists of potential matching medical standard codes, and to output top matching medical standard codes based on the comparison;

enrich the unstructured input with the top matching medical standard codes, and output the unstructured input tagged with the top matching medical standard codes as one of the plurality of input medical texts received by the initialization circuitry.

17. The diagnostic device of claim 16 , wherein the tagging circuitry is further configured to generate a single list of top matching medical standard code from the plurality of logical components.